1. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 1 Are the Means of Several Groups Equal? Ho:Ha: Consider the following two sets of boxplots:

2. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 2 How Different Are They? The key to our test will be thinking about the variation between groups. If the null hypothesis (Ho) is true, all the treatment means estimate the same underlying mean. The means we get for the groups would then vary around the common mean only from natural sampling variation. So, we could act as though the treatment means were just observations and find their variance.

3. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 3 The Ruler Within We have an estimate from the variation within groups. That’s traditionally called the error mean square and written MSE. It’s just the variance of the residuals. Because it’s a pooled variance, we write it We’ve got a separate estimate from the variation between the groups. At least we expect it to estimate if we assume the null hypothesis is true. We call this quantity the treatment mean square (MST).

4. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 4 The F-Statistic When the null hypothesis is true and the treatment means are equal, both MS E and MS T estimate  2, and their ratio should be close to 1. We can use their ratio MS T /MS E to test the null hypothesis: If the treatment means really are different, the numerator will tend to be larger than the denominator, and the ratio will be bigger than 1.

5. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 5 The F-Statistic (cont.) The sampling distribution model for this ratio, found by Sir Ronald Fisher, is called the F-distribution. We call the ratio MS T /MS E the F-statistic. By comparing the F-statistic to the appropriate F-distribution, we (or the computer/calculator) can get a P-value.

6. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 6 The F-Statistic (cont.) The test is one-tailed, because a larger difference in the treatments ultimately leads to a larger F-statistic. So the test is significant if the F-ratio is “big enough” (and the P-value “small enough”). The entire analysis is called Analysis of Variance, commonly abbreviated ANOVA.

7. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 7 The F-Statistic (cont.) Just like Student’s t, the F-models are a family of distributions. However, since we have two variance estimates, MS T and MS E, we have two degrees of freedom parameters. MS T estimates the variance of the treatment means and has k – 1 degrees of freedom when there are k groups. MS E is the pooled estimate of the variance within groups. If there are n observations in each of the k groups, MS E has k(n – 1) degrees of freedom.

8. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 8 The ANOVA Table You’ll often see the Mean Squares and other information put into a table called the ANOVA table. For the soap example in the book, the ANOVA table is:

9. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 9 The F-Table (cont.) Here’s an excerpt from an F-table for  = 0.05:

10. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 10 Assumptions and Conditions As in regression, we must perform our checks of assumptions and conditions in order. And, as in regression, displays of the residuals are often a good way to check the conditions for ANOVA.

11. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 11 Plot the Data… First examine side-by-side boxplots of the data comparing the responses for all of the groups. Check for outliers within any of the groups (and correct them if there are errors in the data). Get an idea of whether the groups have similar spreads (as we’ll need). Get an idea of whether the centers seem to be alike (as the null hypothesis claims) or different. If the individual boxplots are all skewed in the same direction, consider re-expressing the response variable to make them more symmetric.

12. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 12 Independence Assumptions The groups must be independent of each other. No test can verify this assumption—you have to think about how the data were collected. The data within each treatment group must be independent as well. Check the Randomization Condition: Were the data collected with suitable randomization (a representative random sample or random assignment to treatment groups)?

13. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 13 Equal Variance Assumption ANOVA requires that the variances of the treatment groups be equal. To check this assumption, we can check that the groups have similar variances: Similar Variance Condition: Look at side-by-side boxplots of the groups to see whether they have roughly the same spread. Look at the original boxplots of the response values again—in general, do the spreads seem to change systematically with the centers? (This is more of a problem than random differences in spread among the groups and should not be ignored.)

14. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 14 Equal Variance Assumption (cont.) Similar Variance Condition: Look at the residuals plotted against the predicted values. (Larger predicted values lead to larger magnitude residuals, indicating that the condition is violated.)

15. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 15 Equal Variance Assumption (cont.) In our example, neither of the following plots shows a violation of the equal variance assumption:

16. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 16 Normal Population Assumption The F-test requires the underlying errors to follow a Normal Model. We will check a corresponding Nearly Normal Condition: examine a histogram of a Normal probability plot of all the residuals together.

17. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 17 What Can Go Wrong? Watch out for outliers. One outlier in a group can influence the entire F-test and analysis. Watch out for changing variances. If the conditions on the residuals are violated, it may be necessary to re-express the response variable to closer approximate the necessary conditions.

18. Copyright © 2006 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Slide 28- 18 What have we learned? We can compare the means of more than two independent groups based on samples drawn from those groups. We can test the hypothesis that all the means are equal using Analysis of Variance (ANOVA). And, as usual, there are conditions to check. When we want to compare pairs of means (after finding significance with an ANOVA F-test), we need to use a multiple comparisons method.